Estimation of a semiparametric natural direct effect model incorporating baseline covariates

E. J.Tchetgen Tchetgen, I. Shpitser

Research output: Contribution to journalArticle

Abstract

Establishing cause-effect relationships is a standard goal of empirical science. Once the existence of a causal relationship is established, the precise causal mechanism involved becomes a topic of interest. A particularly popular type of mechanism analysis concerns questions of mediation, i.e., to what extent an effect is direct, and to what extent it is mediated by a third variable. A semiparametric theory has recently been proposed that allows multiply robust estimation of direct and mediated marginal effect functionals in observational studies (Tchetgen Tchetgen & Shpitser, 2012). In this paper we extend the theory to handle parametric models of natural direct and indirect effects within levels of pre-exposure variables with an identity or log link function, where the model for the observed data likelihood is otherwise unrestricted. We show that estimation is generally infeasible in such a model because of the curse of dimensionality associated with the required estimation of auxiliary conditional densities or expectations, given high-dimensional covariates. Thus, we consider multiply robust estimation and propose a more general model which assumes that a subset, but not the entirety, of several working models holds.

Original languageEnglish (US)
Pages (from-to)849-864
Number of pages16
JournalBiometrika
Volume101
Issue number4
DOIs
StatePublished - Dec 1 2014
Externally publishedYes

Fingerprint

Direct Effect
Observational Studies
Covariates
Baseline
Robust Estimation
Multiplication
Link Function
Observational Study
Conditional Density
Mediation
Curse of Dimensionality
Conditional Expectation
Parametric Model
Model
observational studies
Likelihood
High-dimensional
Subset
Direct effect
Relationships

Keywords

  • Local efficiency
  • Mediation
  • Multiple robustness
  • Natural direct effect
  • Natural indirect effect

ASJC Scopus subject areas

  • Statistics and Probability
  • Mathematics(all)
  • Agricultural and Biological Sciences (miscellaneous)
  • Agricultural and Biological Sciences(all)
  • Statistics, Probability and Uncertainty
  • Applied Mathematics

Cite this

Estimation of a semiparametric natural direct effect model incorporating baseline covariates. / Tchetgen, E. J.Tchetgen; Shpitser, I.

In: Biometrika, Vol. 101, No. 4, 01.12.2014, p. 849-864.

Research output: Contribution to journalArticle

Tchetgen, E. J.Tchetgen ; Shpitser, I. / Estimation of a semiparametric natural direct effect model incorporating baseline covariates. In: Biometrika. 2014 ; Vol. 101, No. 4. pp. 849-864.
@article{6d5a7988eaff4c47a636491a2e786b91,
title = "Estimation of a semiparametric natural direct effect model incorporating baseline covariates",
abstract = "Establishing cause-effect relationships is a standard goal of empirical science. Once the existence of a causal relationship is established, the precise causal mechanism involved becomes a topic of interest. A particularly popular type of mechanism analysis concerns questions of mediation, i.e., to what extent an effect is direct, and to what extent it is mediated by a third variable. A semiparametric theory has recently been proposed that allows multiply robust estimation of direct and mediated marginal effect functionals in observational studies (Tchetgen Tchetgen & Shpitser, 2012). In this paper we extend the theory to handle parametric models of natural direct and indirect effects within levels of pre-exposure variables with an identity or log link function, where the model for the observed data likelihood is otherwise unrestricted. We show that estimation is generally infeasible in such a model because of the curse of dimensionality associated with the required estimation of auxiliary conditional densities or expectations, given high-dimensional covariates. Thus, we consider multiply robust estimation and propose a more general model which assumes that a subset, but not the entirety, of several working models holds.",
keywords = "Local efficiency, Mediation, Multiple robustness, Natural direct effect, Natural indirect effect",
author = "Tchetgen, {E. J.Tchetgen} and I. Shpitser",
year = "2014",
month = "12",
day = "1",
doi = "10.1093/biomet/asu044",
language = "English (US)",
volume = "101",
pages = "849--864",
journal = "Biometrika",
issn = "0006-3444",
publisher = "Oxford University Press",
number = "4",

}

TY - JOUR

T1 - Estimation of a semiparametric natural direct effect model incorporating baseline covariates

AU - Tchetgen, E. J.Tchetgen

AU - Shpitser, I.

PY - 2014/12/1

Y1 - 2014/12/1

N2 - Establishing cause-effect relationships is a standard goal of empirical science. Once the existence of a causal relationship is established, the precise causal mechanism involved becomes a topic of interest. A particularly popular type of mechanism analysis concerns questions of mediation, i.e., to what extent an effect is direct, and to what extent it is mediated by a third variable. A semiparametric theory has recently been proposed that allows multiply robust estimation of direct and mediated marginal effect functionals in observational studies (Tchetgen Tchetgen & Shpitser, 2012). In this paper we extend the theory to handle parametric models of natural direct and indirect effects within levels of pre-exposure variables with an identity or log link function, where the model for the observed data likelihood is otherwise unrestricted. We show that estimation is generally infeasible in such a model because of the curse of dimensionality associated with the required estimation of auxiliary conditional densities or expectations, given high-dimensional covariates. Thus, we consider multiply robust estimation and propose a more general model which assumes that a subset, but not the entirety, of several working models holds.

AB - Establishing cause-effect relationships is a standard goal of empirical science. Once the existence of a causal relationship is established, the precise causal mechanism involved becomes a topic of interest. A particularly popular type of mechanism analysis concerns questions of mediation, i.e., to what extent an effect is direct, and to what extent it is mediated by a third variable. A semiparametric theory has recently been proposed that allows multiply robust estimation of direct and mediated marginal effect functionals in observational studies (Tchetgen Tchetgen & Shpitser, 2012). In this paper we extend the theory to handle parametric models of natural direct and indirect effects within levels of pre-exposure variables with an identity or log link function, where the model for the observed data likelihood is otherwise unrestricted. We show that estimation is generally infeasible in such a model because of the curse of dimensionality associated with the required estimation of auxiliary conditional densities or expectations, given high-dimensional covariates. Thus, we consider multiply robust estimation and propose a more general model which assumes that a subset, but not the entirety, of several working models holds.

KW - Local efficiency

KW - Mediation

KW - Multiple robustness

KW - Natural direct effect

KW - Natural indirect effect

UR - http://www.scopus.com/inward/record.url?scp=84985992213&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84985992213&partnerID=8YFLogxK

U2 - 10.1093/biomet/asu044

DO - 10.1093/biomet/asu044

M3 - Article

VL - 101

SP - 849

EP - 864

JO - Biometrika

JF - Biometrika

SN - 0006-3444

IS - 4

ER -